Tencent released Hy3 preview in 2026, the first model from its rebuilt Hunyuan AI platform, reaching public availability in under three months after a major infrastructure overhaul in early 2026. The fast-and-slow-thinking fused MoE model features 295 billion total parameters with 21 billion activated, supports 256K context, and scored 74.4% on SWE-bench Verified, trailing Claude Opus 4.6 (80.8%) and GPT-5.4 (78.6%). Tencent deployed the model across six products including Yuanbao and Tencent Docs before launch and is offering it at RMB 1.2 per million input tokens with a two-week free access period. The company positions this as a first step to gather developer feedback while continuing to scale pre-training and reinforcement learning in parallel.
Sep 29, 2026 · 4 sources
Sep 28, 2026 · 7 sources
Sep 30, 2026 · 7 sources
Sep 30, 2026 · 4 sources
Story comments
Loading comments…